Research On Computation Offloading And Resource Allocation In Heterogeneous Environment

With the rapid development of mobile Internet and Internet of Things,the types of user devices are constantly enriched and new mobile applications are constantly emerging.Computation-intensive applications such as pattern recognition,interactive games,virtual reality,and augmented reality not only consume a large number of computation resource and energy,but also demand ultra-low response delay.However,due to the limited computation and energy resources,user devices could not support these applications efficiently.Mobile cloud computing allows users to offload computation tasks to cloud servers for processing.Users could leverage the powerful computing capability of cloud servers to make up for the shortage of computation resource.However,the remote transmission of tasks incurs high transmission delay and communication energy consumption.As an improved technique,mobile edge computing(MEC)could provide computing service in close proximity to users through servers at the edge network,which could overcome the drawbacks of mobile cloud computing.Computation offloading procedure involves computation and communication resources,such that unreasonable computation offloading strategy could incur a waste of resource and even performance loss.Therefore,designing reasonable computation offloading and resource allocation strategy through balancing computation and communication overhead is the key to excavate the potential of edge computing.On the background of heterogeneous networks and heterogeneous computing systems,computation demands,computation resources and edge servers become diverse and heterogeneous.As a result,computation offloading and resource allocation become more intricate.On one hand,the available computation resource of servers and the suitability between tasks and servers should be considered while making computation offloading decision.On the other hand,computation demands should be matched to computing capacity of processors in resource allocation procedure.Therefore,how to balance computation resources and channel states,reasonably match users to servers,schedule processors,and jointly allocate communication resource and heterogeneous computation resources are emerging challenges.To address these challenges,this dissertation is devoted to the research on computation offloading and resource allocation strategy in heterogeneous environment,i.e.,heterogeneous networks and heterogeneous computing systems.The main research content and achievements are summarized as follows.1)Aiming at multiple server MEC system with differentiation of server computing capacity,a joint control method of computation offloading decision and transmission power allocation is developed to minimize the system-level energy consumption.In the presence of multiple servers,it is a challenge to achieve the lowest system-level energy consumption while comprehensively considering the available computation resource,channel state and latency requirement.The systemlevel energy consumption is minimized by jointly optimizing computation offloading decision and transmission power allocation.The formulated energy consumption minimization problem falls into the category of mixed integer and nonlinear program.To solve it efficiently,we decompose the original problem into transmission power allocation problem and computation offloading decision problem.As for transmission power allocation problem,the lower bound of feasible transmission power proves to be the optimal solution.As for computation offloading decision problem,by exploring the effect of available computation resource and channel state on the system-level energy consumption,we propose an iterative solution framework to optimize computation offloading decision.Simulation results show that the proposed algorithm has near optimal performance.2)As for MEC system where general servers coexist with dedicated servers,we propose a method to jointly manage computation offloading and resource allocation to minimize the system-level overhead,i.e.,weighted sum of user energy consumption and task execution delay.As the computing system undergoes a trend toward heterogeneity,some edge severs become dedicated,which could accelerate the computation of specific tasks but is incapable of executing other tasks.A new challenge about how to reasonably offload diverse computation tasks to fully utilize the potential of dedicated server should be addressed.We formulate a system-wide overhead minimization problem of joint task offloading decision and resource allocation.Leveraging the structure characteristic of the formulated problem,we decompose the it into resource allocation problem and task offloading decision problem.Resource allocation problem could be addressed by bisection method and convex optimization theory.To reasonably match users to servers,a task offloading decision initialization method is proposed based on the designed user preference function and offloading loss function.A heuristic method is proposed to further optimize the task offloading decision.Simulation results demonstrate that the proposed algorithm performs very closely to the optimal algorithm,and outperforms the other benchmark approaches.3)As for servers based on CPU/GPU processors,we design a joint method of processor scheduling and resource allocation to maximize the server utility,which is determined by the weighted sum of user task processing efficiency and block generation efficiency.In the considered blockchain-enabled MEC system,blockchain tasks together with offloaded tasks are diverse in parallelism demands,while servers consist of diverse processors with vastly different parallel computing power.In view of this phenomenon,we aim to maximize server utility by jointly optimizing processor scheduling and resource allocation.The formulated problem is equivalently transformed into joint resource allocation problem and processor scheduling problem.A suboptimal joint resource allocation strategy could be obtained by leveraging convex optimization theory to alternately solve for heterogeneous computation resources and bandwidth resource allocation strategy.To reasonably schedule tasks on processors,we propose a heuristic method to iteratively optimize processor scheduling strategy.Simulation results verify that the proposed method makes the processor scheduling strategy to the most optimization,and that the proposed joint resource allocation method outperforms the other benchmark methods.